The United States has cleared Nvidia to sell its H200 artificial intelligence chip to roughly 10 Chinese companies, but the approvals have not yet produced deliveries, leaving one of the most closely watched semiconductor trade openings in limbo.

Reuters reported on May 14 that the cleared buyers include major Chinese technology groups such as Alibaba, Tencent, ByteDance and JD.com, citing people familiar with the matter. The report said not a single H200 chip had been delivered so far, even after Washington allowed the sales under a restricted licensing framework. The gap between export approval and physical shipment has become the central issue for Nvidia, Chinese cloud buyers and investors trying to assess whether the U.S.-China AI chip channel is reopening or merely entering a more complex phase of managed trade.

The development puts Nvidia CEO Jensen Huang at the center of a diplomatic and commercial test. Huang has been seeking to revive China sales while preserving Nvidia’s standing with U.S. policymakers who remain concerned that advanced AI processors could strengthen Chinese military, surveillance or strategic computing capabilities. His efforts come as U.S. controls have already altered Nvidia’s access to one of the world’s largest AI infrastructure markets and accelerated China’s efforts to build domestic substitutes.

The H200 is not Nvidia’s newest chip architecture, but it remains a high-end accelerator for generative AI and high-performance computing. Nvidia has described the H200 as a Hopper-based Tensor Core GPU with faster and larger memory designed to improve performance on large language models and other data-intensive workloads. That profile makes the chip strategically important for Chinese internet companies seeking to train, fine-tune and serve AI models at scale, especially in cloud environments where memory bandwidth and GPU availability can determine model economics.

The U.S. decision does not represent a full return to unrestricted semiconductor sales. It follows a January revision by the Commerce Department’s Bureau of Industry and Security, which shifted license review policy for certain advanced computing chips bound for China and Macau from a presumption of denial to case-by-case review when security requirements are met. The change created a legal path for specific exports, but it did not remove Washington’s ability to impose conditions, review end users, restrict diversion and monitor risks tied to military or prohibited uses.

That distinction matters for Nvidia’s investors because approved licenses are not the same as recognized revenue. The Reuters report indicates that Washington’s clearance has not overcome operational and political obstacles. Chinese companies may still face pressure from Beijing to prioritize domestic chips, while U.S. conditions may make the transactions less attractive or more difficult to execute. The result is a rare situation in which the U.S. government has allowed a limited route for sales, but the commercial channel remains blocked by strategic mistrust on both sides.

For Nvidia, the stakes are substantial even if the immediate financial impact is uncertain. The company has dominated the global market for AI accelerators, and its data-center business has become the core driver of its valuation. China historically represented an important market for Nvidia’s advanced computing products, but successive U.S. export restrictions forced the company to redesign products, manage license uncertainty and accept that some of its most powerful chips could not be freely sold into the country.

H200 approvals could help Nvidia defend its platform position in China at a time when local customers are being encouraged to shift toward domestic alternatives. The strategic logic for Nvidia extends beyond chip units sold. Its ecosystem includes CUDA software, networking, systems integration and developer tools that make its GPUs embedded in enterprise and cloud AI workflows. If Chinese companies continue to build around Nvidia hardware, the company preserves influence over AI infrastructure standards. If they move decisively to domestic chips, Nvidia risks losing not only sales but also software and platform gravity in a large AI market.

For Chinese companies, the calculation is similarly complicated. Access to H200 processors would likely improve near-term AI capacity and help cloud providers serve customers developing generative AI applications. Alibaba, Tencent, ByteDance and JD.com all have major data, cloud, commerce or consumer internet operations that can benefit from faster AI inference and training capacity. Yet buying U.S.-approved chips under stringent conditions may conflict with Beijing’s broader objective of reducing reliance on foreign semiconductor supply chains.

China’s industrial policy has increasingly favored domestic AI hardware suppliers as U.S. export controls tightened. That policy is partly defensive: dependence on U.S. chips leaves Chinese companies exposed to future license cancellations, shipment delays and product redesigns. It is also strategic: domestic demand can help Chinese semiconductor companies scale production, improve software stacks and narrow performance gaps. The more Beijing pushes firms toward local alternatives, the harder it becomes for Nvidia to convert U.S. approvals into actual shipments.

Washington’s own policy environment is also unsettled. Export controls on advanced computing chips have become one of the primary instruments of U.S. technology policy toward China. Supporters argue that limits are necessary to prevent sensitive AI capabilities from aiding the Chinese military or state security apparatus. Critics of easing restrictions warn that even older or restricted high-end chips can still materially improve China’s AI capacity if deployed at scale. Business groups and some technology executives counter that a total cutoff may accelerate Chinese self-sufficiency and reduce U.S. leverage over global AI standards.

The H200 case sits between those positions. It suggests Washington is willing to permit some controlled access to U.S. AI hardware, but only under conditions intended to limit strategic leakage. That middle path may be commercially awkward. Too many restrictions can deter buyers, complicate logistics and invite Chinese regulatory resistance. Too few restrictions can trigger backlash in Congress and among national security officials. Nvidia is therefore trying to navigate not a simple export market, but a policy architecture in which each sale can become a geopolitical test.

The specific mechanics of stalled deliveries remain important for the market. Reuters reported that no chips have been shipped despite clearance for around 10 Chinese firms. The absence of shipments means investors cannot yet assume a revenue contribution from the approvals. It also leaves open questions about whether Chinese authorities will permit purchases, whether U.S. agencies will require additional assurances, and whether supply-chain procedures will satisfy both governments.

Those uncertainties are material because AI chip demand is one of the most important themes in global equity markets. Nvidia’s valuation has been supported by expectations that hyperscale cloud operators, sovereign AI projects, enterprise customers and model developers will continue expanding GPU clusters. Any reopening of the Chinese market could add another source of demand, but only if approvals translate into deliverable products and repeatable transactions. A one-time license clearance without shipments would be less meaningful than a stable framework under which Chinese customers can plan procurement.

The H200 itself also illustrates how quickly the AI hardware cycle is moving. While it remains powerful, Nvidia has been pushing newer generations of AI infrastructure, and global customers are already planning around next-generation systems. For China, H200 access could still be valuable because export restrictions have constrained access to the most advanced alternatives. For Nvidia, selling H200s may help monetize inventory and preserve relationships while maintaining compliance with U.S. rules. However, any delay reduces the chip’s commercial window as customers elsewhere move to newer products.

The case also carries implications for Nvidia’s competitors and the broader semiconductor supply chain. Advanced AI chips depend on high-bandwidth memory, sophisticated packaging, foundry capacity, server integration and networking equipment. If H200 shipments to China eventually proceed, suppliers linked to AI servers and memory could benefit indirectly. If the channel remains stalled, Chinese demand may shift more aggressively toward domestic accelerator suppliers and alternative architectures, reinforcing a bifurcation of the AI hardware market.

Cloud competition is another important dimension. U.S. hyperscalers have built AI services around large clusters of Nvidia GPUs, giving them substantial infrastructure advantages. Chinese cloud providers face constraints in obtaining comparable chips, even as domestic demand for generative AI tools, enterprise automation and consumer AI applications continues to rise. H200 approvals could narrow that gap at the margin, but the stalled delivery status means Chinese companies cannot yet count on the capacity. That may affect product road maps, model training schedules and pricing strategies in China’s AI cloud market.

The approvals also highlight a recurring problem for multinational technology companies operating between Washington and Beijing: legal permission from one capital may not be enough. Nvidia must satisfy U.S. export-control rules, reassure American officials, maintain customer relationships in China and avoid becoming a symbol of either strategic concession or foreign dependence. Chinese buyers must balance performance needs against policy pressure to localize procurement. Both sides have incentives to reach a transaction, but neither side wants to concede control over a strategically important technology channel.

For U.S. policymakers, the core question is whether controlled exports preserve American influence or undermine the purpose of export controls. Allowing H200 sales could keep Chinese AI developers tied to U.S. hardware and software ecosystems while generating revenue for an American company. But it could also provide compute resources to entities whose activities Washington may later view as strategically problematic. The effectiveness of the policy depends on enforcement, end-use verification, shipment controls and the willingness to halt sales if conditions are breached.

For Beijing, the decision is whether near-term access to powerful U.S. chips is worth the long-term dependency risk. Chinese companies may want H200 capacity for immediate competitiveness, but Chinese authorities have reasons to discourage reliance on a supply chain that Washington can interrupt. The stalled deliveries suggest that China may be using its own approval process and informal policy signals to shape corporate behavior, even when U.S. licenses have been granted.

The market reaction to the Reuters report reflected the possibility of upside, but the underlying facts support caution. Approval for about 10 firms is a notable policy development. The lack of deliveries is an equally important constraint. Until chips move, customers install them and Nvidia recognizes sales, the episode remains more a signal of potential reopening than evidence of a restored China revenue stream.

The issue is likely to remain a focal point for semiconductor investors because it links three large themes: AI infrastructure demand, U.S.-China strategic competition and Nvidia’s ability to sustain growth across regulated markets. Any evidence that shipments are beginning would likely be interpreted as a positive sign for Nvidia and parts of the AI supply chain. Conversely, continued delays could indicate that export licenses are insufficient to overcome political resistance and that China’s AI hardware market is becoming more structurally separated from U.S. suppliers.

For now, the U.S. clearance of Nvidia H200 sales to selected Chinese firms is best understood as a partial opening, not a completed breakthrough. Washington has created a pathway; Nvidia has potential customers; Chinese technology groups have strong technical incentives to buy. But deliveries remain stalled, and the commercial outcome depends on whether the two governments can tolerate a controlled flow of advanced AI hardware across a strategic fault line that neither side fully trusts.